Fedora 33's "Enterprise Linux Next" Effort Approved - Testbed For Raising CPU Requirements, Etc

[qarium]

my conclusion on reading all this is...
a lowest target would be
SSE4.2 Westmere

or if this cheap/low-end it not par of their business modell.
Ryzen1000 with Zen1.0 means AVX2 would be a good start.
i have 2 systems right now a threadripper 1920 and 1900 and they are Zen1.0 means AVX2

lets face it they doing business and their sales on ultra cheap hardware like SSE4.2 Westmere is maybe zero?...

but ryzen means AVX2 is really a big part of the business.

[coder]

Would you be running RHEL9 on Atom CPUs though... Buying cheap, low performance microprocessors for the type of server workload you'd pay a yearly subscription fee for isn't particularly common.

You have no idea. These are sold in up to 16-core configurations, in order to address a range of specialized use cases.

People associate Atom with low-end, but it's also Intel's most power-efficient product line. For a long time, now, Intel has reserved the Atom branding exclusively for the embedded/server products based on their power-efficient cores.

[coder]

I don't think Intel Atoms are a real showstopper. For my opinion they have never been really good at anything, instead they were crippled and just low-cost solutions and sold via their price. Even efficiency isn't that great. There is not much energy comsumption turned into abysmal performance...

These are not your father's Atom CPUs. They're based on Intel's Goldmont core, which has IPC somewhere between Core2 and Nehalem, while sipping just a couple W of power per core.

However, the Tremont cores are really next-level. And still doesn't even support vanilla AVX! So, now we're looking at a whole new generation of non-AVX CPUs. Thank Intel, not me.

Let's go on: RHEL isn't exactly lowcost, so it's most of the time not paired with cheapo crippled hardware.

If you're running an Atom-based server, it's probably not for cost reasons.

I still don't mind.

You don't have to. Behemoth dinosaurs probably didn't mind not being more mammal or bird-like, until they got wiped out.

The main point is to stop thinking about baseline ISA requirements as the solution to utilizing CPU features. It's archaic thinking. The energy is much better spent on something akin to SPIR-V for CPUs.

[nils_]

And the city of Munich ditched Windows for Linux, but had encountered a number of practical issues to the point they decided to move back to Windows.

The Linux desktop does exist, but not so much in the real corporate world for all sorts of pragmatic reasons.

With Munich it's more of a political decision.

[programmerjake]

The energy is much better spent on something akin to SPIR-V for CPUs.

WebAssembly anyone?

[coder]

WebAssembly anyone?

I don't know if it's the best intermediate representation or API abstraction for the purpose, but it could at least be a step in the right direction.

I'm willing to reserve judgement until we at least see how those efforts pan out.

I wonder if advancement on this front won't be needed to properly utilize the capabilities of any given RISC-V implementation.

[brouhaha]

Would you be running RHEL9 on Atom CPUs though... Buying cheap, low performance microprocessors for the type of server workload you'd pay a yearly subscription fee for isn't particularly common.

Not all servers run compute-heavy workloads.

I run multiple SuperMicro severs using the Atom Denverton C3758 processors (Goldmont microarchitecture). They are optimized for low power consumption, and have decent enough performance. They even have some crypto offload intended for server use, though at the moment I have to disable that in the BIOS because it interferes with dm-crypt. :-(

The C3000 series CPUs are explicitly supported by RHEL8.

The next generation of server CPUs in this family is likely to use the Temont microarchitecture and still not support AVX.